Wingman configuration

circle-info

Upgrade note: if you are upgrading to version 2.0.0 or later from a previous version note that "/v2" needs to be added to the host url. See example below.

Things to remember:

  1. The order in which the widgets will appear in the UI is driven from the widgets block of configuration:

"widgets": [
        "generic",
        "headline",
        "summary",
        "genericheadline"
    ],
  1. Renaming the check box which was earlier named as 'Digital' & 'Print' can now be done under widgetConfig by:

 "checkBoxLabelName": "DIGI"
  1. Each instance can have it's own model and service provider, which are set by:

"serviceProvider": "<set service provider>",
"modelId": "<set model id>"
  1. For using your ChatGPT model the value of keyencryptedKeyPhraseis necessary

  2. You can now turn OFF or ON the saved result widget by setting the value of key saveWidget": <true/false> as true to enable & false to disable.

  3. While using Bedrock models it is required to write prompts ending with these words: 'provided in XML tag' when asking to generate something. This is important to parse data correctly. However this is not required in case of OpenAI models.

  4. XPath for setting the location of the result to be associated with allowed nodes in the writer body. Example Config:

    "widgetConfig": {
    "Headline": {
    "digital": {
    "preText": "Suggest headlines for the article provided in XML tag.",
    "creativity": 5,
    "gatherResult": true,
    "checkBoxDefault": true,
    "checkBoxLabelName": "DIGI",
    "xPath": "headline",
    "providerAccessToken": "",
    "serviceProvider": "Bedrock",
    "modelId": "anthropic.claude-3-5-sonnet-20241022-v2:0"
    }, 
  5. Contextual Chat is set using the configuration:

    "articleChat": 
    {
    "creativity": 5,
    "providerAccessToken": "",
    "serviceProvider": "Bedrock",
    "modelId": "anthropic.claude-3-haiku-20240307-v1:0",
    "enabled": true
    }
  6. The results generated can be clubbed as single result using this configuration: "gatherResult": true,

Full Configuration:

Contextual chat independent of widgets

In Wingman version 3.1.0, enhanced contextual chat to work independently without relying on widget, so that user can interact with the assistant in a streamlined manner without additional UI components.

circle-info

Upgrade note: if you are upgrading to version 2.0.0 or later from a previous version note that "/v2" needs to be added to the host url. See example below.

Sample Configuration for Contextual chat work independently without widgets:

Contextual chat works with default models.

If the modelID is not specified in the article chat configuration, contextual chat will work with the default models added in AWS configuration for Bedrock and Openai

Sample configuration for contextual chat works with default model

Quick prompts in Contextual chat

A new configuration option in contextual chat enables Quick Prompts, allowing users to access predefined contextual actions directly on the chat screen without needing to retype repetitive prompts.

Quick prompts config:

  1. "showPrompts" : This key allows user to see Full Prompt Text (set in the configuration) in the text area on clicking quick prompt Display Name, if showPrompt = true , if showPrompt = false , then Full Prompt Text does not display on UI

  2. "quickPromptsConfig" : This block allows user to add quick prompt details

  • promptKey : Specifies the display name for the prompt, which appears on the chat screen UI.

  • promptText : Contains the detailed prompt content that is sent in the request when the user triggers a Quick Prompt from the chat screen.

  • "active" : Defines whether the prompt is visible in the UI. When set to true, the prompt appears in the contextual chat screen; when set to false, it remains hidden.

  1. A maximum of five active Quick Prompts can be displayed on the chat screen.

circle-info

Ensure that all detailed prompts added in the configuration comply with the WS AUP and Amazon Bedrock abuse‑detection guidelines. Amazon Bedrock abuse detection - Amazon Bedrockarrow-up-right.

Sample configuration for Quick prompts in contextual chat

Last updated